Machine Learning Surrogates for Optimizing Transportation Policies with Agent-Based Models

TL;DR

This paper tackles the high computational cost of evaluating transportation policies with agent-based models by proposing Graph Neural Network surrogates trained on MATSim Paris data. The authors formulate a regression problem to predict edge-level changes in car volume under policy interventions, employing a dual-graph representation and a hybrid GNN architecture with PointNet, Transformer, and GAT layers, optimized via Mean Squared Error. Results show the surrogate can predict policy-induced traffic changes quickly (≈0.1 s per scenario) and with substantial accuracy (overall $R^2$ ≈ 0.76), particularly for roads directly impacted by policies. The work demonstrates potential for large-scale simulation-based optimization and real-time scenario evaluation, and outlines future steps for multimodal predictions, dynamic scenarios, and cross-city transferability.

Abstract

Rapid urbanization and growing urban populations worldwide present significant challenges for cities, including increased traffic congestion and air pollution. Effective strategies are needed to manage traffic volumes and reduce emissions. In practice, traditional traffic flow simulations are used to test those strategies. However, high computational intensity usually limits their applicability in investigating a magnitude of different scenarios to evaluate best policies. This paper presents a first approach of using Graph Neural Networks (GNN) as surrogates for large-scale agent-based simulation models. In a case study using the MATSim model of Paris, the GNN effectively learned the impacts of capacity reduction policies on citywide traffic flow. Performance analysis across various road types and scenarios revealed that the GNN could accurately capture policy-induced effects on edge-based traffic volumes, particularly on roads directly affected by the policies and those with higher traffic volumes.

Figures (2)

• Figure 1: Architecture of the Graph Neural Network
• Figure 2: Randomly chosen test data: Actual (left) and predicted (right) change in car volume